An automatic Tensorflow-CUDA-Docker-Jupyter machine on Google Cloud Platform


For a class I'm teaching (on deep learning and art) I had to create a machine that auto starts a jupyter notebook with tensorflow and GPU support. Just create an instance and presto - Jupyter notebook with TF and GPU!
How awesome is that?

Well... building it wasn't that simple.
So for your enjoyment - here's my recipe:
Continue reading "An automatic Tensorflow-CUDA-Docker-Jupyter machine on Google Cloud Platform"

Share

Projector-Camera Calibration - the "easy" way

First let me open by saying projector-camera calibration is NOT EASY. But it's technically not complicated too.

It is however, an amalgamation of optimizations that accrue and accumulate error with each step, so that the end product is not far from a random guess.
So 3D reconstructions I was able to get from my calibrated pro-cam were just a distorted mess of points.

Nevertheless, here come the deets.
Continue reading "Projector-Camera Calibration - the "easy" way"

Share

Revisiting graph-cut segmentation with SLIC and color histograms [w/Python]

As part of the computer vision class I'm teaching at SBU I asked students to implement a segmentation method based on SLIC superpixels. Here is my boilerplate implementation.

This follows the work I've done a very long time ago (2010) on the same subject.

For graph-cut I've used PyMaxflow: https://github.com/pmneila/PyMaxflow, which is very easily installed by just pip install PyMaxflow

The method is simple:

  • Calculate SLIC superpixels (the SKImage implementation)
  • Use markings to determine the foreground and background color histograms (from the superpixels under the markings)
  • Setup a graph with a straightforward energy model: Smoothness term = K-L-Div between superpix histogram and neighbor superpix histogram, and Match term = inf if marked as BG or FG, or K-L-Div between SuperPix histogram and FG and BG.
  • To find neighbors I've used Delaunay tessellation (from scipy.spatial), for simplicity. But a full neighbor finding could be implemented by looking at all the neighbors on the superpix's boundary.
  • Color histograms are 2D over H-S (from the HSV)

Result

Share

[Python] OpenCV capturing from a v4l2 device

I tried to set the capture format on a webcam from OpenCV's cv2.VideoCapture and ran into a problem: it's using the wrong IOCTL command.
So I used python-v4l2capture to get images from the device, which allows more control.
Here is the gist:

Enjoy!
Roy

Share

OpenCV Python YAML persistance

I wasn't able to find online a complete example on how to persist OpenCV matrices in Python (so really NumPy arrays) to YAML like what cv::FileStorage will give you.

So here's a short snippet:

import numpy as np
import yaml

# A yaml constructor is for loading from a yaml node.
# This is taken from: http://stackoverflow.com/a/15942429
def opencv_matrix_constructor(loader, node):
    mapping = loader.construct_mapping(node, deep=True)
    mat = np.array(mapping["data"])
    mat.resize(mapping["rows"], mapping["cols"])
    return mat
yaml.add_constructor(u"tag:yaml.org,2002:opencv-matrix", opencv_matrix_constructor)

# A yaml representer is for dumping structs into a yaml node.
# So for an opencv_matrix type (to be compatible with c++'s FileStorage) we save the rows, cols, type and flattened-data
def opencv_matrix_representer(dumper, mat):
    mapping = {'rows': mat.shape[0], 'cols': mat.shape[1], 'dt': 'd', 'data': mat.reshape(-1).tolist()}
    return dumper.represent_mapping(u"tag:yaml.org,2002:opencv-matrix", mapping)
yaml.add_representer(np.ndarray, opencv_matrix_representer)


#example
with open('output.yaml', 'w') as f:
    f.write("%YAML:1.0")
    yaml.dump({"a matrix": np.zeros((10,10)), "another_one": np.zeros((2,4))}, f)

#   a matrix: !!opencv-matrix
#     cols: 10
#     data: [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
#       0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
#       0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
#       0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
#       0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
#       0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
#       0.0, 0.0, 0.0, 0.0, 0.0]
#     dt: d
#     rows: 10
#   another_one: !!opencv-matrix
#     cols: 4
#     data: [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]
#     dt: d
#     rows: 2

with open('output.yaml', 'r') as f:
    print yaml.load(f)
  
#  {'a matrix': array([[ 0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.],
#         [ 0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.],
#         [ 0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.],
#         [ 0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.],
#         [ 0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.],
#         [ 0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.],
#         [ 0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.],
#         [ 0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.],
#         [ 0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.],
#         [ 0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.]]), 'another_one': array([[ 0.,  0.,  0.,  0.],
#         [ 0.,  0.,  0.,  0.]])}

There you go

Share

GDoc/LaTeX compilation GUI with Tkinter/Python [w/ code]

ยกHola mis amigos!
I'm learning spanish, but I'm also annoyed with collaborating on LaTeX papers. That's why I've created the GDoc-LaTeXifier so the syntax will be clear when I collaborate on a paper with a remote friend.
But now we both want to compile a PDF on our machines. So I've created the tiny shell script that downloads the paper and runs PDFLaTeX.
The problem is that this opens a new terminal window and runs the script. I've been able to sort it out so that it closes the terminal window when it's done, by on my friend's mac it doesn't, so he ends up with a ton of open windows.
Enter - the GDoc/LaTeX compiler GUI.
Continue reading "GDoc/LaTeX compilation GUI with Tkinter/Python [w/ code]"

Share

Getting all the links from a MediaWiki format using PyParsing

Hi,
Just sharing a snippet of code. Part of a project I'm doing, I need to analyse the links in the Wikipedia corpus. While using the API is one solution, it doesn't retain the order of where links appear in the page. It also returns links that are not part of the main text, which makes the linkage DB very cluttered.
So, I set out to parse the raw MediaWiki format all Wikipedia articles are written in, to get only the relevant links and in order. I call them contextual because they live inside the text and have context.
Initially I used string matching, and other complex string scraping parsing methods. It was a bust. There are too many end-cases to deal with. That is when I discovered PyParsing, the excellent parsing library. It did the job, and here are the results.
Continue reading "Getting all the links from a MediaWiki format using PyParsing"

Share

Download all your Last.fm loved tracks in two simple steps

I'm a fan of Last.fm online radio, and I have a habit of marking every good song that I hear as a "loved track". Over the years I got quite a list, and so I decided to turn it into my jogging playlist. But for that, I need all the songs downloaded to my computer so I can put them on my mobile. While Last.fm does link to Amazon for downloading all the loved songs for pay, I'm going to walk the fine moral line here and suggest how you can download every song from existing free YouTube videos.
If it really bothers you, think of it as if I created a YouTube playlist and now I'm using my data plan to stream the songs off YT itself..
Moral issues resolved, we can move on to the scripting.

Update (4/27/12): youtube-dl.py has moved: https://github.com/rg3/youtube-dl/, and also added a very neat --extract-audio option so you can get the songs in audio right away (it basically does a conversion in a second step).
Continue reading "Download all your Last.fm loved tracks in two simple steps"

Share

10 lines-of-code OCR HTTP service with Python, Tesseract and Tornado

Hi

I believe that every builder-hacker should have their own little Swiss-army-knife server that just does everything they need, but as a webservice. You can basically do anything as a service nowadays: image/audio/video manipulation, mock-cloud data storage, offload heavy computation, and so on.
Tornado, the lightweight Python webserver is perfect for this, and since so many of the projects these days have Python binding (see python-tesseract), it should be a breeze to integrate them with minimal work.
Let's see how it's done

Continue reading "10 lines-of-code OCR HTTP service with Python, Tesseract and Tornado"

Share